People like a person who speaks well of them, even if they know that person is a flatterer. Similarly, a computer that provides flattering messages is preferred to a computer that provides generic feedback. Here we examined whether the type of computer message can affect a user's behavior. According to the norm of reciprocity (i.e., people tend to accept the suggestions and requests of people who have been good to them), we expected, in turn, that a user would tend to accept the suggestions and requests of a computer that also spoke well of them. In the first experiment, 72 participants played a quiz game with a computer. The computer sometimes suggested that the participant's answer might be wrong; as such, the number of times the participant changed his or her mind was recorded. In the second experiment, 60 participants played a "20-question" game wherein the computer guessed the name of an animal the participant was asked to think on. In both experiments, participants received one of three types of computer messages — positive (flattering), negative (criticizing), or neutral (generic) — during the task. At the end, the computer asked the participants to update its database as much as they were willing to. Contrary to the prediction, the flattering computer was not appreciated very highly. However, it made the user calmer and more relaxed. The criticizing computer was rated as less likable than the other computers. The type of messages did not influence the acceptance rate of the computer's suggestion and request. The results suggest that the contents of computer messages can affect users' feeling but this effect may not be strong enough to change their behavior.